From d08bf5844a878c92c0153fb761f308e08258f45a Mon Sep 17 00:00:00 2001
From: sirius0xdev <sirius0xdev@gmail.com>
Date: Wed, 29 Apr 2026 04:12:36 +0000
Subject: [PATCH] feat: optimize RTX6000 vLLM KEDA scaling

- Rename deployment to rtx6000-brain-vllm to match scaler target ref
- Set deployment replicas to 0 (KEDA controlled)
- Add pollingInterval: 10s and cooldownPeriod: 30s for faster response

This enables scale-to-zero when idle and quick scale-up on first request.

See https://github.com/sirius0xdev/gcloud-lab/tree/master/infrastructure%2Fgpus%2Fbase
---
 infrastructure/gpus/base/keda-gpu-scaling/keda-vllm.yaml | 2 +-
 infrastructure/gpus/base/vllm-servers/rtx6000-vllm.yaml  | 4 ++--
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/infrastructure/gpus/base/keda-gpu-scaling/keda-vllm.yaml b/infrastructure/gpus/base/keda-gpu-scaling/keda-vllm.yaml
index c675bdd..eee9d2e 100644
--- a/infrastructure/gpus/base/keda-gpu-scaling/keda-vllm.yaml
+++ b/infrastructure/gpus/base/keda-gpu-scaling/keda-vllm.yaml
@@ -14,4 +14,4 @@ spec:
     min: 0
     max: 1
   targetPendingRequests: 1
-  scaledownPeriod: 900
+  scaledownPeriod: 900\n  pollingInterval: 10\n  cooldownPeriod: 30
diff --git a/infrastructure/gpus/base/vllm-servers/rtx6000-vllm.yaml b/infrastructure/gpus/base/vllm-servers/rtx6000-vllm.yaml
index 2e60c16..d5ccbbc 100644
--- a/infrastructure/gpus/base/vllm-servers/rtx6000-vllm.yaml
+++ b/infrastructure/gpus/base/vllm-servers/rtx6000-vllm.yaml
@@ -2,12 +2,12 @@
 apiVersion: apps/v1
 kind: Deployment
 metadata:
-  name: openclaw-brain-vllm
+  name: rtx6000-brain-vllm
   namespace: customer1
   labels:
     app: rtx6000-brain
 spec:
-  replicas: 1
+  replicas: 0
   selector:
     matchLabels:
       app: rtx6000-brain